MultiDTI

MultiDTI predicts drug-target interactions by applying multi-modal representation learning to combine heterogeneous network associations, drug/target sequence data, and chemical-structure information to enable DTI predictions for novel chemical entities.

Key Features:

• Multi-Modal Representation Learning: Integrates interaction and association information from heterogeneous networks with drug and target sequence data to form comprehensive representations.
• Joint Learning Framework: Maps drugs, targets, side effects, and disease nodes into a common representational space to relate entities across modalities.
• Handling Novel Chemical Entities: Projects new chemical entities based on their chemical structures into the learned common space to enable predictions outside pre-existing networks.
• Predictive Performance: Validated by 10-fold cross-validation with reported AUC-ROC of 0.961 and AUC-PR of 0.947.
• Empirical Corroboration: Some predicted interactions have been corroborated by the ChEMBL database.

Scientific Applications:

• Drug Development: Facilitates identification of potential new drug targets and candidate interactions for preclinical assessment.
• Side Effect Analysis: Supports analysis of associations between drugs and side effects through integrated network and sequence representations.
• Drug Repositioning: Aids discovery of alternative therapeutic uses for existing compounds by predicting novel DTIs.

Methodology:

Combines heterogeneous network interaction/association data with drug/target sequence data via multi-modal representation learning; employs a joint learning framework to embed drugs, targets, side effects, and disease nodes into a common space; maps chemical structures of novel compounds into that space for DTI prediction; evaluates performance with 10-fold cross-validation reporting AUC-ROC 0.961 and AUC-PR 0.947 and compares predicted interactions against ChEMBL.